mixup: Beyond Empirical Risk Minimization
This addresses issues of generalization and robustness in deep learning for practitioners, offering a novel regularization technique.
The paper tackles the problem of undesirable behaviors in large deep neural networks, such as memorization and sensitivity to adversarial examples, by proposing mixup, a simple learning principle that trains on convex combinations of pairs of examples and labels, resulting in improved generalization across multiple datasets including ImageNet-2012 and CIFAR-10/100.
Large deep neural networks are powerful, but exhibit undesirable behaviors such as memorization and sensitivity to adversarial examples. In this work, we propose mixup, a simple learning principle to alleviate these issues. In essence, mixup trains a neural network on convex combinations of pairs of examples and their labels. By doing so, mixup regularizes the neural network to favor simple linear behavior in-between training examples. Our experiments on the ImageNet-2012, CIFAR-10, CIFAR-100, Google commands and UCI datasets show that mixup improves the generalization of state-of-the-art neural network architectures. We also find that mixup reduces the memorization of corrupt labels, increases the robustness to adversarial examples, and stabilizes the training of generative adversarial networks.